Hong Kong SFC: AI adoption is becoming increasingly widespread, and responsible development at scale remains challenging.

The Hong Kong Securities and Futures Commission (SFC) published its first Survey on the Use and Preparedness of Artificial Intelligence by Large-Scale Institutions. The results show that institutions have widely adopted AI across business and monitoring functions. The report examines institutions' use of AI in business activities and operations, the degree of integration of AI with internal systems, and institutions' plans for further AI adoption. It also explores how institutions are adjusting their governance and risk management arrangements in response to increasingly widespread AI use, as well as the challenges involved in deploying AI responsibly. The survey was conducted between July and August 2026 and covered 20 large financial institutions and their 78 affiliated licensed corporations in Hong Kong. Responses show that AI use is permeating business and monitoring functions, integration with internal systems is increasing, and some applications are becoming more autonomous. The survey findings particularly note the growing prevalence of agentic AI, used mainly for information retrieval, content preparation, software development, operational workflows and certain trading-related activities. 30% of surveyed institutions have deployed agentic AI in their business operations, while another 25% are running pilots, involving varying degrees of autonomy and human intervention. Other key survey findings include: AI adoption is gaining momentum, with 95% of surveyed institutions expecting to accelerate AI deployment over the next 12 months. Adoption now spans front-, middle- and back-office activities, and system integration continues to deepen, with 90% of institutions having connected AI tools to internal systems. Institutions are using AI for cyber defence, with 90% adopting or assessing the use of frontier AI models for this purpose, helping to speed up analysis and expand detection coverage. Practical difficulties in further broadening AI adoption include technical integration, moving pilot projects into full production, uncertainty over costs and business value, data limitations, and skills shortages. As applications proliferate, institutions face challenges in expanding the scope of governance and monitoring measures while maintaining clear accountability arrangements and human oversight. The survey findings will help the SFC formulate its AI regulatory approach, which has four key focus areas: (i) protecting investor interests as AI reshapes how investors interact with institutions; (ii) strengthening cybersecurity and operational resilience to address new threats and dependencies arising from AI; (iii) re-examining how to implement human accountability and human oversight in increasingly autonomous workflows to ensure effective AI governance; and (iv) building knowledge and capability through exchange and collaboration. Dr Eric Yip, Executive Director of the SFC's Intermediaries Division, said: "We adopt a safety-first, human-centric and collaboration-led regulatory approach. As AI becomes increasingly autonomous, we must ensure that human primacy is not undermined." He added: "Driving AI adoption at scale requires more than powerful models and promising use cases; it also requires reliable infrastructure, robust data, appropriate expertise, and governance systems that keep pace with technological developments." The SFC reminds industry institutions to comply with applicable legal and regulatory requirements to support the responsible use of AI in Hong Kong's capital markets. The SFC will focus on ensuring that human accountability is maintained throughout business processes that use AI, including by setting clear objectives, clearly defining the scope of data access and authorised actions, and establishing effective monitoring and intervention mechanisms. The relevant controls should be commensurate with the permissions and potential impact of the applications, rather than requiring human approval of every output or action.